A lot of this is still ideas floating around my skull and various notes (electronic or paper), not fully organised.
Ello is, as I've just noted in a post there, very search-resistant, which is becoming a major pain-point. My post index (see my profile page there) is very sadly out of date, though it has a few items on it.
https://ello.co/dredmorbius/post/dlz9c2z6x-tvkd7rxtxwpw
I'd put some thoughts together on Reddit and community and what seems to work, or not:
https://www.reddit.com/r/dredmorbius/comments/20yhxc/reddit_...
In the spirit of Thomas Edison, a long list of things that don't work:
https://www.reddit.com/r/dredmorbius/comments/1w73g4/on_comm...
You've found two of the better items I'd put together on trust and dialectic discussion at Ello.
Though not directly related to size, I've also suggested standardising topical "channels" of discussion:
https://ello.co/dredmorbius/post/ftpx7lwnfjtopygu5xrkhg
Somewhere amongst my writing is a suggestion for a tiered set of vaguely Dunbar's Number sized network clusters. Say, 50-150 people per group, themselves organised into clusters of perhaps 50 groups themselves. A small level of tiers gets you to current global population size (5-7 levels as I recall). If each group surfaces, say, 10 - 100 items per day of choice public content, and the filtering is applied upwards, then ... maybe there's a way to have both local and global discussion.
Topical interests might be overlays on that, based on net interest. A topic with a 1:50 interest level would draw from all 50 neighbouring clusters for its own base level. A topic with a 1:2500 interest would scale to two levels up -- this retains some level of "locality" whilst also allowing for broader discussion.
And branching out between clusters would be possible, but effectively with some additional cost -- UI/UX-imposed or other. Not large, but present.
There's an existing (or was) network that was organised somewhat on these terms, though I believe it's not done much. Building initial interest is difficult.